If anybody has any comments on beta features which were not submitted to premier support or listed on the survey, then let us know.
Features in Intel MKL 10.1 Beta 2 release:
- BLAS performance has improved on 32- and 64-bit Intel Xeon processors as well as on the forthcoming Intel micro-architecture code named Nehalem. Among the functions optimized are SGEMM, DGEMM, CGEMM, ZGEMM, and DTRSM.
- The performance of the direct sparse solver (DSS/PARDISO) has been improved by 35% on average. In addition, support for separate backward and forward substitution, a parameter for checking sparse structure, and means to turn off iterative refinement (DSS interface) have been added.
- The capability to track the progress of a lengthy computation and/or interrupt the computation has been added via a callback function mechanism.
- Transposition functions have been added to Intel MKL.
- The C++ std::complex type can now be used instead of MKL-specific complex types.
- An implementation of the Boost uBLAS matrix-matrix multiplication routine is now provided which will make use of the highly optimized version of DGEMM in the Intel MKL BLAS.
- Routines for computing the sum and product of two sparse matrices have been added as well as support for the single precision, complex, and double complex data types.
- The Cumulative Normal Distribution (CdfNorm), its inverse (CdfNormInv), and the inverse complementary error function (ErfcInv) of the Vector Math Library (VML) have been optimized to achieve much improved performance from the implementations in Beta 1.
- Many of the VML and Vector Statistical Library (VSL) functions have now been optimized on the Intel micro-architecture code named Nehalem achieving up to 67% improvement over previous versions.
- On Linux*: Eclipse IDE Infopop support for VML functions and VSL service functions in the CDT 5.0 version.
- On Windows*: Restrictions to creating DLLs from static libraries have been removed.
A few changes to note:
- The configuration file functionality has been removed. See the user guide for alternative means to configure the behavior of Intel MKL.
- When functions in Intel MKL are called from an MPI program they will be run on 1 thread by default (i.e., in the absence of explicit controls).